1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Procedures for interfacing to the RTAS on CHRP machines.
6 * Peter Bergner, IBM March 2001.
7 * Copyright (C) 2001 IBM.
10 #define pr_fmt(fmt) "rtas: " fmt
12 #include <linux/bsearch.h>
13 #include <linux/capability.h>
14 #include <linux/delay.h>
15 #include <linux/export.h>
16 #include <linux/init.h>
17 #include <linux/kconfig.h>
18 #include <linux/kernel.h>
19 #include <linux/lockdep.h>
20 #include <linux/memblock.h>
21 #include <linux/mutex.h>
23 #include <linux/of_fdt.h>
24 #include <linux/reboot.h>
25 #include <linux/sched.h>
26 #include <linux/security.h>
27 #include <linux/slab.h>
28 #include <linux/spinlock.h>
29 #include <linux/stdarg.h>
30 #include <linux/syscalls.h>
31 #include <linux/types.h>
32 #include <linux/uaccess.h>
33 #include <linux/xarray.h>
35 #include <asm/delay.h>
36 #include <asm/firmware.h>
37 #include <asm/interrupt.h>
38 #include <asm/machdep.h>
41 #include <asm/rtas-work-area.h>
44 #include <asm/trace.h>
48 /* Indexes into the args buffer, -1 if not used */
54 * Assumed buffer size per the spec if the function does not
55 * have a size parameter, e.g. ibm,errinjct. 0 if unused.
61 * struct rtas_function - Descriptor for RTAS functions.
63 * @token: Value of @name if it exists under the /rtas node.
64 * @name: Function name.
65 * @filter: If non-NULL, invoking this function via the rtas syscall is
66 * generally allowed, and @filter describes constraints on the
67 * arguments. See also @banned_for_syscall_on_le.
68 * @banned_for_syscall_on_le: Set when call via sys_rtas is generally allowed
69 * but specifically restricted on ppc64le. Such
70 * functions are believed to have no users on
71 * ppc64le, and we want to keep it that way. It does
72 * not make sense for this to be set when @filter
74 * @lock: Pointer to an optional dedicated per-function mutex. This
75 * should be set for functions that require multiple calls in
76 * sequence to complete a single operation, and such sequences
77 * will disrupt each other if allowed to interleave. Users of
78 * this function are required to hold the associated lock for
79 * the duration of the call sequence. Add an explanatory
80 * comment to the function table entry if setting this member.
82 struct rtas_function {
84 const bool banned_for_syscall_on_le:1;
85 const char * const name;
86 const struct rtas_filter *filter;
91 * Per-function locks for sequence-based RTAS functions.
93 static DEFINE_MUTEX(rtas_ibm_activate_firmware_lock);
94 static DEFINE_MUTEX(rtas_ibm_get_dynamic_sensor_state_lock);
95 static DEFINE_MUTEX(rtas_ibm_get_indices_lock);
96 static DEFINE_MUTEX(rtas_ibm_lpar_perftools_lock);
97 static DEFINE_MUTEX(rtas_ibm_physical_attestation_lock);
98 static DEFINE_MUTEX(rtas_ibm_set_dynamic_indicator_lock);
99 DEFINE_MUTEX(rtas_ibm_get_vpd_lock);
101 static struct rtas_function rtas_function_table[] __ro_after_init = {
102 [RTAS_FNIDX__CHECK_EXCEPTION] = {
103 .name = "check-exception",
105 [RTAS_FNIDX__DISPLAY_CHARACTER] = {
106 .name = "display-character",
107 .filter = &(const struct rtas_filter) {
108 .buf_idx1 = -1, .size_idx1 = -1,
109 .buf_idx2 = -1, .size_idx2 = -1,
112 [RTAS_FNIDX__EVENT_SCAN] = {
113 .name = "event-scan",
115 [RTAS_FNIDX__FREEZE_TIME_BASE] = {
116 .name = "freeze-time-base",
118 [RTAS_FNIDX__GET_POWER_LEVEL] = {
119 .name = "get-power-level",
120 .filter = &(const struct rtas_filter) {
121 .buf_idx1 = -1, .size_idx1 = -1,
122 .buf_idx2 = -1, .size_idx2 = -1,
125 [RTAS_FNIDX__GET_SENSOR_STATE] = {
126 .name = "get-sensor-state",
127 .filter = &(const struct rtas_filter) {
128 .buf_idx1 = -1, .size_idx1 = -1,
129 .buf_idx2 = -1, .size_idx2 = -1,
132 [RTAS_FNIDX__GET_TERM_CHAR] = {
133 .name = "get-term-char",
135 [RTAS_FNIDX__GET_TIME_OF_DAY] = {
136 .name = "get-time-of-day",
137 .filter = &(const struct rtas_filter) {
138 .buf_idx1 = -1, .size_idx1 = -1,
139 .buf_idx2 = -1, .size_idx2 = -1,
142 [RTAS_FNIDX__IBM_ACTIVATE_FIRMWARE] = {
143 .name = "ibm,activate-firmware",
144 .filter = &(const struct rtas_filter) {
145 .buf_idx1 = -1, .size_idx1 = -1,
146 .buf_idx2 = -1, .size_idx2 = -1,
149 * PAPR+ as of v2.13 doesn't explicitly impose any
150 * restriction, but this typically requires multiple
151 * calls before success, and there's no reason to
152 * allow sequences to interleave.
154 .lock = &rtas_ibm_activate_firmware_lock,
156 [RTAS_FNIDX__IBM_CBE_START_PTCAL] = {
157 .name = "ibm,cbe-start-ptcal",
159 [RTAS_FNIDX__IBM_CBE_STOP_PTCAL] = {
160 .name = "ibm,cbe-stop-ptcal",
162 [RTAS_FNIDX__IBM_CHANGE_MSI] = {
163 .name = "ibm,change-msi",
165 [RTAS_FNIDX__IBM_CLOSE_ERRINJCT] = {
166 .name = "ibm,close-errinjct",
167 .filter = &(const struct rtas_filter) {
168 .buf_idx1 = -1, .size_idx1 = -1,
169 .buf_idx2 = -1, .size_idx2 = -1,
172 [RTAS_FNIDX__IBM_CONFIGURE_BRIDGE] = {
173 .name = "ibm,configure-bridge",
175 [RTAS_FNIDX__IBM_CONFIGURE_CONNECTOR] = {
176 .name = "ibm,configure-connector",
177 .filter = &(const struct rtas_filter) {
178 .buf_idx1 = 0, .size_idx1 = -1,
179 .buf_idx2 = 1, .size_idx2 = -1,
183 [RTAS_FNIDX__IBM_CONFIGURE_KERNEL_DUMP] = {
184 .name = "ibm,configure-kernel-dump",
186 [RTAS_FNIDX__IBM_CONFIGURE_PE] = {
187 .name = "ibm,configure-pe",
189 [RTAS_FNIDX__IBM_CREATE_PE_DMA_WINDOW] = {
190 .name = "ibm,create-pe-dma-window",
192 [RTAS_FNIDX__IBM_DISPLAY_MESSAGE] = {
193 .name = "ibm,display-message",
194 .filter = &(const struct rtas_filter) {
195 .buf_idx1 = 0, .size_idx1 = -1,
196 .buf_idx2 = -1, .size_idx2 = -1,
199 [RTAS_FNIDX__IBM_ERRINJCT] = {
200 .name = "ibm,errinjct",
201 .filter = &(const struct rtas_filter) {
202 .buf_idx1 = 2, .size_idx1 = -1,
203 .buf_idx2 = -1, .size_idx2 = -1,
207 [RTAS_FNIDX__IBM_EXTI2C] = {
208 .name = "ibm,exti2c",
210 [RTAS_FNIDX__IBM_GET_CONFIG_ADDR_INFO] = {
211 .name = "ibm,get-config-addr-info",
213 [RTAS_FNIDX__IBM_GET_CONFIG_ADDR_INFO2] = {
214 .name = "ibm,get-config-addr-info2",
215 .filter = &(const struct rtas_filter) {
216 .buf_idx1 = -1, .size_idx1 = -1,
217 .buf_idx2 = -1, .size_idx2 = -1,
220 [RTAS_FNIDX__IBM_GET_DYNAMIC_SENSOR_STATE] = {
221 .name = "ibm,get-dynamic-sensor-state",
222 .filter = &(const struct rtas_filter) {
223 .buf_idx1 = 1, .size_idx1 = -1,
224 .buf_idx2 = -1, .size_idx2 = -1,
227 * PAPR+ v2.13 R1–7.3.19–3 is explicit that the OS
228 * must not call ibm,get-dynamic-sensor-state with
229 * different inputs until a non-retry status has been
232 .lock = &rtas_ibm_get_dynamic_sensor_state_lock,
234 [RTAS_FNIDX__IBM_GET_INDICES] = {
235 .name = "ibm,get-indices",
236 .filter = &(const struct rtas_filter) {
237 .buf_idx1 = 2, .size_idx1 = 3,
238 .buf_idx2 = -1, .size_idx2 = -1,
241 * PAPR+ v2.13 R1–7.3.17–2 says that the OS must not
242 * interleave ibm,get-indices call sequences with
245 .lock = &rtas_ibm_get_indices_lock,
247 [RTAS_FNIDX__IBM_GET_RIO_TOPOLOGY] = {
248 .name = "ibm,get-rio-topology",
250 [RTAS_FNIDX__IBM_GET_SYSTEM_PARAMETER] = {
251 .name = "ibm,get-system-parameter",
252 .filter = &(const struct rtas_filter) {
253 .buf_idx1 = 1, .size_idx1 = 2,
254 .buf_idx2 = -1, .size_idx2 = -1,
257 [RTAS_FNIDX__IBM_GET_VPD] = {
258 .name = "ibm,get-vpd",
259 .filter = &(const struct rtas_filter) {
260 .buf_idx1 = 0, .size_idx1 = -1,
261 .buf_idx2 = 1, .size_idx2 = 2,
264 * PAPR+ v2.13 R1–7.3.20–4 indicates that sequences
265 * should not be allowed to interleave.
267 .lock = &rtas_ibm_get_vpd_lock,
269 [RTAS_FNIDX__IBM_GET_XIVE] = {
270 .name = "ibm,get-xive",
272 [RTAS_FNIDX__IBM_INT_OFF] = {
273 .name = "ibm,int-off",
275 [RTAS_FNIDX__IBM_INT_ON] = {
276 .name = "ibm,int-on",
278 [RTAS_FNIDX__IBM_IO_QUIESCE_ACK] = {
279 .name = "ibm,io-quiesce-ack",
281 [RTAS_FNIDX__IBM_LPAR_PERFTOOLS] = {
282 .name = "ibm,lpar-perftools",
283 .filter = &(const struct rtas_filter) {
284 .buf_idx1 = 2, .size_idx1 = 3,
285 .buf_idx2 = -1, .size_idx2 = -1,
288 * PAPR+ v2.13 R1–7.3.26–6 says the OS should allow
289 * only one call sequence in progress at a time.
291 .lock = &rtas_ibm_lpar_perftools_lock,
293 [RTAS_FNIDX__IBM_MANAGE_FLASH_IMAGE] = {
294 .name = "ibm,manage-flash-image",
296 [RTAS_FNIDX__IBM_MANAGE_STORAGE_PRESERVATION] = {
297 .name = "ibm,manage-storage-preservation",
299 [RTAS_FNIDX__IBM_NMI_INTERLOCK] = {
300 .name = "ibm,nmi-interlock",
302 [RTAS_FNIDX__IBM_NMI_REGISTER] = {
303 .name = "ibm,nmi-register",
305 [RTAS_FNIDX__IBM_OPEN_ERRINJCT] = {
306 .name = "ibm,open-errinjct",
307 .filter = &(const struct rtas_filter) {
308 .buf_idx1 = -1, .size_idx1 = -1,
309 .buf_idx2 = -1, .size_idx2 = -1,
312 [RTAS_FNIDX__IBM_OPEN_SRIOV_ALLOW_UNFREEZE] = {
313 .name = "ibm,open-sriov-allow-unfreeze",
315 [RTAS_FNIDX__IBM_OPEN_SRIOV_MAP_PE_NUMBER] = {
316 .name = "ibm,open-sriov-map-pe-number",
318 [RTAS_FNIDX__IBM_OS_TERM] = {
319 .name = "ibm,os-term",
321 [RTAS_FNIDX__IBM_PARTNER_CONTROL] = {
322 .name = "ibm,partner-control",
324 [RTAS_FNIDX__IBM_PHYSICAL_ATTESTATION] = {
325 .name = "ibm,physical-attestation",
326 .filter = &(const struct rtas_filter) {
327 .buf_idx1 = 0, .size_idx1 = 1,
328 .buf_idx2 = -1, .size_idx2 = -1,
331 * This follows a sequence-based pattern similar to
332 * ibm,get-vpd et al. Since PAPR+ restricts
333 * interleaving call sequences for other functions of
334 * this style, assume the restriction applies here,
335 * even though it's not explicit in the spec.
337 .lock = &rtas_ibm_physical_attestation_lock,
339 [RTAS_FNIDX__IBM_PLATFORM_DUMP] = {
340 .name = "ibm,platform-dump",
341 .filter = &(const struct rtas_filter) {
342 .buf_idx1 = 4, .size_idx1 = 5,
343 .buf_idx2 = -1, .size_idx2 = -1,
346 * PAPR+ v2.13 7.3.3.4.1 indicates that concurrent
347 * sequences of ibm,platform-dump are allowed if they
348 * are operating on different dump tags. So leave the
349 * lock pointer unset for now. This may need
350 * reconsideration if kernel-internal users appear.
353 [RTAS_FNIDX__IBM_POWER_OFF_UPS] = {
354 .name = "ibm,power-off-ups",
356 [RTAS_FNIDX__IBM_QUERY_INTERRUPT_SOURCE_NUMBER] = {
357 .name = "ibm,query-interrupt-source-number",
359 [RTAS_FNIDX__IBM_QUERY_PE_DMA_WINDOW] = {
360 .name = "ibm,query-pe-dma-window",
362 [RTAS_FNIDX__IBM_READ_PCI_CONFIG] = {
363 .name = "ibm,read-pci-config",
365 [RTAS_FNIDX__IBM_READ_SLOT_RESET_STATE] = {
366 .name = "ibm,read-slot-reset-state",
367 .filter = &(const struct rtas_filter) {
368 .buf_idx1 = -1, .size_idx1 = -1,
369 .buf_idx2 = -1, .size_idx2 = -1,
372 [RTAS_FNIDX__IBM_READ_SLOT_RESET_STATE2] = {
373 .name = "ibm,read-slot-reset-state2",
375 [RTAS_FNIDX__IBM_REMOVE_PE_DMA_WINDOW] = {
376 .name = "ibm,remove-pe-dma-window",
378 [RTAS_FNIDX__IBM_RESET_PE_DMA_WINDOW] = {
380 * Note: PAPR+ v2.13 7.3.31.4.1 spells this as
381 * "ibm,reset-pe-dma-windows" (plural), but RTAS
382 * implementations use the singular form in practice.
384 .name = "ibm,reset-pe-dma-window",
386 [RTAS_FNIDX__IBM_SCAN_LOG_DUMP] = {
387 .name = "ibm,scan-log-dump",
388 .filter = &(const struct rtas_filter) {
389 .buf_idx1 = 0, .size_idx1 = 1,
390 .buf_idx2 = -1, .size_idx2 = -1,
393 [RTAS_FNIDX__IBM_SET_DYNAMIC_INDICATOR] = {
394 .name = "ibm,set-dynamic-indicator",
395 .filter = &(const struct rtas_filter) {
396 .buf_idx1 = 2, .size_idx1 = -1,
397 .buf_idx2 = -1, .size_idx2 = -1,
400 * PAPR+ v2.13 R1–7.3.18–3 says the OS must not call
401 * this function with different inputs until a
402 * non-retry status has been returned.
404 .lock = &rtas_ibm_set_dynamic_indicator_lock,
406 [RTAS_FNIDX__IBM_SET_EEH_OPTION] = {
407 .name = "ibm,set-eeh-option",
408 .filter = &(const struct rtas_filter) {
409 .buf_idx1 = -1, .size_idx1 = -1,
410 .buf_idx2 = -1, .size_idx2 = -1,
413 [RTAS_FNIDX__IBM_SET_SLOT_RESET] = {
414 .name = "ibm,set-slot-reset",
416 [RTAS_FNIDX__IBM_SET_SYSTEM_PARAMETER] = {
417 .name = "ibm,set-system-parameter",
418 .filter = &(const struct rtas_filter) {
419 .buf_idx1 = 1, .size_idx1 = -1,
420 .buf_idx2 = -1, .size_idx2 = -1,
423 [RTAS_FNIDX__IBM_SET_XIVE] = {
424 .name = "ibm,set-xive",
426 [RTAS_FNIDX__IBM_SLOT_ERROR_DETAIL] = {
427 .name = "ibm,slot-error-detail",
429 [RTAS_FNIDX__IBM_SUSPEND_ME] = {
430 .name = "ibm,suspend-me",
431 .banned_for_syscall_on_le = true,
432 .filter = &(const struct rtas_filter) {
433 .buf_idx1 = -1, .size_idx1 = -1,
434 .buf_idx2 = -1, .size_idx2 = -1,
437 [RTAS_FNIDX__IBM_TUNE_DMA_PARMS] = {
438 .name = "ibm,tune-dma-parms",
440 [RTAS_FNIDX__IBM_UPDATE_FLASH_64_AND_REBOOT] = {
441 .name = "ibm,update-flash-64-and-reboot",
443 [RTAS_FNIDX__IBM_UPDATE_NODES] = {
444 .name = "ibm,update-nodes",
445 .banned_for_syscall_on_le = true,
446 .filter = &(const struct rtas_filter) {
447 .buf_idx1 = 0, .size_idx1 = -1,
448 .buf_idx2 = -1, .size_idx2 = -1,
452 [RTAS_FNIDX__IBM_UPDATE_PROPERTIES] = {
453 .name = "ibm,update-properties",
454 .banned_for_syscall_on_le = true,
455 .filter = &(const struct rtas_filter) {
456 .buf_idx1 = 0, .size_idx1 = -1,
457 .buf_idx2 = -1, .size_idx2 = -1,
461 [RTAS_FNIDX__IBM_VALIDATE_FLASH_IMAGE] = {
462 .name = "ibm,validate-flash-image",
464 [RTAS_FNIDX__IBM_WRITE_PCI_CONFIG] = {
465 .name = "ibm,write-pci-config",
467 [RTAS_FNIDX__NVRAM_FETCH] = {
468 .name = "nvram-fetch",
470 [RTAS_FNIDX__NVRAM_STORE] = {
471 .name = "nvram-store",
473 [RTAS_FNIDX__POWER_OFF] = {
476 [RTAS_FNIDX__PUT_TERM_CHAR] = {
477 .name = "put-term-char",
479 [RTAS_FNIDX__QUERY_CPU_STOPPED_STATE] = {
480 .name = "query-cpu-stopped-state",
482 [RTAS_FNIDX__READ_PCI_CONFIG] = {
483 .name = "read-pci-config",
485 [RTAS_FNIDX__RTAS_LAST_ERROR] = {
486 .name = "rtas-last-error",
488 [RTAS_FNIDX__SET_INDICATOR] = {
489 .name = "set-indicator",
490 .filter = &(const struct rtas_filter) {
491 .buf_idx1 = -1, .size_idx1 = -1,
492 .buf_idx2 = -1, .size_idx2 = -1,
495 [RTAS_FNIDX__SET_POWER_LEVEL] = {
496 .name = "set-power-level",
497 .filter = &(const struct rtas_filter) {
498 .buf_idx1 = -1, .size_idx1 = -1,
499 .buf_idx2 = -1, .size_idx2 = -1,
502 [RTAS_FNIDX__SET_TIME_FOR_POWER_ON] = {
503 .name = "set-time-for-power-on",
504 .filter = &(const struct rtas_filter) {
505 .buf_idx1 = -1, .size_idx1 = -1,
506 .buf_idx2 = -1, .size_idx2 = -1,
509 [RTAS_FNIDX__SET_TIME_OF_DAY] = {
510 .name = "set-time-of-day",
511 .filter = &(const struct rtas_filter) {
512 .buf_idx1 = -1, .size_idx1 = -1,
513 .buf_idx2 = -1, .size_idx2 = -1,
516 [RTAS_FNIDX__START_CPU] = {
519 [RTAS_FNIDX__STOP_SELF] = {
522 [RTAS_FNIDX__SYSTEM_REBOOT] = {
523 .name = "system-reboot",
525 [RTAS_FNIDX__THAW_TIME_BASE] = {
526 .name = "thaw-time-base",
528 [RTAS_FNIDX__WRITE_PCI_CONFIG] = {
529 .name = "write-pci-config",
533 #define for_each_rtas_function(funcp) \
534 for (funcp = &rtas_function_table[0]; \
535 funcp < &rtas_function_table[ARRAY_SIZE(rtas_function_table)]; \
539 * Nearly all RTAS calls need to be serialized. All uses of the
540 * default rtas_args block must hold rtas_lock.
542 * Exceptions to the RTAS serialization requirement (e.g. stop-self)
543 * must use a separate rtas_args structure.
545 static DEFINE_RAW_SPINLOCK(rtas_lock);
546 static struct rtas_args rtas_args;
549 * rtas_function_token() - RTAS function token lookup.
550 * @handle: Function handle, e.g. RTAS_FN_EVENT_SCAN.
552 * Context: Any context.
553 * Return: the token value for the function if implemented by this platform,
554 * otherwise RTAS_UNKNOWN_SERVICE.
556 s32 rtas_function_token(const rtas_fn_handle_t handle)
558 const size_t index = handle.index;
559 const bool out_of_bounds = index >= ARRAY_SIZE(rtas_function_table);
561 if (WARN_ONCE(out_of_bounds, "invalid function index %zu", index))
562 return RTAS_UNKNOWN_SERVICE;
564 * Various drivers attempt token lookups on non-RTAS
568 return RTAS_UNKNOWN_SERVICE;
570 return rtas_function_table[index].token;
572 EXPORT_SYMBOL_GPL(rtas_function_token);
574 static int rtas_function_cmp(const void *a, const void *b)
576 const struct rtas_function *f1 = a;
577 const struct rtas_function *f2 = b;
579 return strcmp(f1->name, f2->name);
583 * Boot-time initialization of the function table needs the lookup to
584 * return a non-const-qualified object. Use rtas_name_to_function()
585 * in all other contexts.
587 static struct rtas_function *__rtas_name_to_function(const char *name)
589 const struct rtas_function key = {
592 struct rtas_function *found;
594 found = bsearch(&key, rtas_function_table, ARRAY_SIZE(rtas_function_table),
595 sizeof(rtas_function_table[0]), rtas_function_cmp);
600 static const struct rtas_function *rtas_name_to_function(const char *name)
602 return __rtas_name_to_function(name);
605 static DEFINE_XARRAY(rtas_token_to_function_xarray);
607 static int __init rtas_token_to_function_xarray_init(void)
609 const struct rtas_function *func;
612 for_each_rtas_function(func) {
613 const s32 token = func->token;
615 if (token == RTAS_UNKNOWN_SERVICE)
618 err = xa_err(xa_store(&rtas_token_to_function_xarray,
619 token, (void *)func, GFP_KERNEL));
626 arch_initcall(rtas_token_to_function_xarray_init);
629 * For use by sys_rtas(), where the token value is provided by user
630 * space and we don't want to warn on failed lookups.
632 static const struct rtas_function *rtas_token_to_function_untrusted(s32 token)
634 return xa_load(&rtas_token_to_function_xarray, token);
638 * Reverse lookup for deriving the function descriptor from a
639 * known-good token value in contexts where the former is not already
640 * available. @token must be valid, e.g. derived from the result of a
641 * prior lookup against the function table.
643 static const struct rtas_function *rtas_token_to_function(s32 token)
645 const struct rtas_function *func;
647 if (WARN_ONCE(token < 0, "invalid token %d", token))
650 func = rtas_token_to_function_untrusted(token);
654 * Fall back to linear scan in case the reverse mapping hasn't
655 * been initialized yet.
657 if (xa_empty(&rtas_token_to_function_xarray)) {
658 for_each_rtas_function(func) {
659 if (func->token == token)
664 WARN_ONCE(true, "unexpected failed lookup for token %d", token);
668 /* This is here deliberately so it's only used in this file */
669 void enter_rtas(unsigned long);
671 static void __do_enter_rtas(struct rtas_args *args)
673 enter_rtas(__pa(args));
674 srr_regs_clobbered(); /* rtas uses SRRs, invalidate */
677 static void __do_enter_rtas_trace(struct rtas_args *args)
679 const struct rtas_function *func = rtas_token_to_function(be32_to_cpu(args->token));
682 * If there is a per-function lock, it must be held by the
686 lockdep_assert_held(func->lock);
688 if (args == &rtas_args)
689 lockdep_assert_held(&rtas_lock);
691 trace_rtas_input(args, func->name);
692 trace_rtas_ll_entry(args);
694 __do_enter_rtas(args);
696 trace_rtas_ll_exit(args);
697 trace_rtas_output(args, func->name);
700 static void do_enter_rtas(struct rtas_args *args)
702 const unsigned long msr = mfmsr();
704 * Situations where we want to skip any active tracepoints for
707 * 1. The last code executed on an offline CPU as it stops,
708 * i.e. we're about to call stop-self. The tracepoints'
709 * function name lookup uses xarray, which uses RCU, which
710 * isn't valid to call on an offline CPU. Any events
711 * emitted on an offline CPU will be discarded anyway.
713 * 2. In real mode, as when invoking ibm,nmi-interlock from
714 * the pseries MCE handler. We cannot count on trace
715 * buffers or the entries in rtas_token_to_function_xarray
716 * to be contained in the RMO.
718 const unsigned long mask = MSR_IR | MSR_DR;
719 const bool can_trace = likely(cpu_online(raw_smp_processor_id()) &&
720 (msr & mask) == mask);
722 * Make sure MSR[RI] is currently enabled as it will be forced later
725 BUG_ON(!(msr & MSR_RI));
727 BUG_ON(!irqs_disabled());
729 hard_irq_disable(); /* Ensure MSR[EE] is disabled on PPC64 */
732 __do_enter_rtas_trace(args);
734 __do_enter_rtas(args);
739 DEFINE_SPINLOCK(rtas_data_buf_lock);
740 EXPORT_SYMBOL_GPL(rtas_data_buf_lock);
742 char rtas_data_buf[RTAS_DATA_BUF_SIZE] __aligned(SZ_4K);
743 EXPORT_SYMBOL_GPL(rtas_data_buf);
745 unsigned long rtas_rmo_buf;
748 * If non-NULL, this gets called when the kernel terminates.
749 * This is done like this so rtas_flash can be a module.
751 void (*rtas_flash_term_hook)(int);
752 EXPORT_SYMBOL_GPL(rtas_flash_term_hook);
755 * call_rtas_display_status and call_rtas_display_status_delay
756 * are designed only for very early low-level debugging, which
757 * is why the token is hard-coded to 10.
759 static void call_rtas_display_status(unsigned char c)
766 raw_spin_lock_irqsave(&rtas_lock, flags);
767 rtas_call_unlocked(&rtas_args, 10, 1, 1, NULL, c);
768 raw_spin_unlock_irqrestore(&rtas_lock, flags);
771 static void call_rtas_display_status_delay(char c)
773 static int pending_newline = 0; /* did last write end with unprinted newline? */
774 static int width = 16;
778 call_rtas_display_status(' ');
783 if (pending_newline) {
784 call_rtas_display_status('\r');
785 call_rtas_display_status('\n');
789 call_rtas_display_status(c);
795 void __init udbg_init_rtas_panel(void)
797 udbg_putc = call_rtas_display_status_delay;
800 #ifdef CONFIG_UDBG_RTAS_CONSOLE
802 /* If you think you're dying before early_init_dt_scan_rtas() does its
803 * work, you can hard code the token values for your firmware here and
804 * hardcode rtas.base/entry etc.
806 static unsigned int rtas_putchar_token = RTAS_UNKNOWN_SERVICE;
807 static unsigned int rtas_getchar_token = RTAS_UNKNOWN_SERVICE;
809 static void udbg_rtascon_putc(char c)
816 /* Add CRs before LFs */
818 udbg_rtascon_putc('\r');
820 /* if there is more than one character to be displayed, wait a bit */
821 for (tries = 0; tries < 16; tries++) {
822 if (rtas_call(rtas_putchar_token, 1, 1, NULL, c) == 0)
828 static int udbg_rtascon_getc_poll(void)
835 if (rtas_call(rtas_getchar_token, 0, 2, &c))
841 static int udbg_rtascon_getc(void)
845 while ((c = udbg_rtascon_getc_poll()) == -1)
852 void __init udbg_init_rtas_console(void)
854 udbg_putc = udbg_rtascon_putc;
855 udbg_getc = udbg_rtascon_getc;
856 udbg_getc_poll = udbg_rtascon_getc_poll;
858 #endif /* CONFIG_UDBG_RTAS_CONSOLE */
860 void rtas_progress(char *s, unsigned short hex)
862 struct device_node *root;
866 static int display_character, set_indicator;
867 static int display_width, display_lines, form_feed;
868 static const int *row_width;
869 static DEFINE_SPINLOCK(progress_lock);
870 static int current_line;
871 static int pending_newline = 0; /* did last write end with unprinted newline? */
876 if (display_width == 0) {
877 display_width = 0x10;
878 if ((root = of_find_node_by_path("/rtas"))) {
879 if ((p = of_get_property(root,
880 "ibm,display-line-length", NULL)))
881 display_width = be32_to_cpu(*p);
882 if ((p = of_get_property(root,
883 "ibm,form-feed", NULL)))
884 form_feed = be32_to_cpu(*p);
885 if ((p = of_get_property(root,
886 "ibm,display-number-of-lines", NULL)))
887 display_lines = be32_to_cpu(*p);
888 row_width = of_get_property(root,
889 "ibm,display-truncation-length", NULL);
892 display_character = rtas_function_token(RTAS_FN_DISPLAY_CHARACTER);
893 set_indicator = rtas_function_token(RTAS_FN_SET_INDICATOR);
896 if (display_character == RTAS_UNKNOWN_SERVICE) {
897 /* use hex display if available */
898 if (set_indicator != RTAS_UNKNOWN_SERVICE)
899 rtas_call(set_indicator, 3, 1, NULL, 6, 0, hex);
903 spin_lock(&progress_lock);
906 * Last write ended with newline, but we didn't print it since
907 * it would just clear the bottom line of output. Print it now
910 * If no newline is pending and form feed is supported, clear the
911 * display with a form feed; otherwise, print a CR to start output
912 * at the beginning of the line.
914 if (pending_newline) {
915 rtas_call(display_character, 1, 1, NULL, '\r');
916 rtas_call(display_character, 1, 1, NULL, '\n');
921 rtas_call(display_character, 1, 1, NULL,
924 rtas_call(display_character, 1, 1, NULL, '\r');
928 width = row_width[current_line];
930 width = display_width;
933 if (*os == '\n' || *os == '\r') {
934 /* If newline is the last character, save it
935 * until next call to avoid bumping up the
938 if (*os == '\n' && !os[1]) {
941 if (current_line > display_lines-1)
942 current_line = display_lines-1;
943 spin_unlock(&progress_lock);
947 /* RTAS wants CR-LF, not just LF */
950 rtas_call(display_character, 1, 1, NULL, '\r');
951 rtas_call(display_character, 1, 1, NULL, '\n');
953 /* CR might be used to re-draw a line, so we'll
954 * leave it alone and not add LF.
956 rtas_call(display_character, 1, 1, NULL, *os);
960 width = row_width[current_line];
962 width = display_width;
965 rtas_call(display_character, 1, 1, NULL, *os);
970 /* if we overwrite the screen length */
972 while ((*os != 0) && (*os != '\n') && (*os != '\r'))
976 spin_unlock(&progress_lock);
978 EXPORT_SYMBOL_GPL(rtas_progress); /* needed by rtas_flash module */
980 int rtas_token(const char *service)
982 const struct rtas_function *func;
985 if (rtas.dev == NULL)
986 return RTAS_UNKNOWN_SERVICE;
988 func = rtas_name_to_function(service);
992 * The caller is looking up a name that is not known to be an
993 * RTAS function. Either it's a function that needs to be
994 * added to the table, or they're misusing rtas_token() to
995 * access non-function properties of the /rtas node. Warn and
996 * fall back to the legacy behavior.
998 WARN_ONCE(1, "unknown function `%s`, should it be added to rtas_function_table?\n",
1001 tokp = of_get_property(rtas.dev, service, NULL);
1002 return tokp ? be32_to_cpu(*tokp) : RTAS_UNKNOWN_SERVICE;
1004 EXPORT_SYMBOL_GPL(rtas_token);
1006 #ifdef CONFIG_RTAS_ERROR_LOGGING
1008 static u32 rtas_error_log_max __ro_after_init = RTAS_ERROR_LOG_MAX;
1011 * Return the firmware-specified size of the error log buffer
1012 * for all rtas calls that require an error buffer argument.
1013 * This includes 'check-exception' and 'rtas-last-error'.
1015 int rtas_get_error_log_max(void)
1017 return rtas_error_log_max;
1020 static void __init init_error_log_max(void)
1022 static const char propname[] __initconst = "rtas-error-log-max";
1025 if (of_property_read_u32(rtas.dev, propname, &max)) {
1026 pr_warn("%s not found, using default of %u\n",
1027 propname, RTAS_ERROR_LOG_MAX);
1028 max = RTAS_ERROR_LOG_MAX;
1031 if (max > RTAS_ERROR_LOG_MAX) {
1032 pr_warn("%s = %u, clamping max error log size to %u\n",
1033 propname, max, RTAS_ERROR_LOG_MAX);
1034 max = RTAS_ERROR_LOG_MAX;
1037 rtas_error_log_max = max;
1041 static char rtas_err_buf[RTAS_ERROR_LOG_MAX];
1043 /** Return a copy of the detailed error text associated with the
1044 * most recent failed call to rtas. Because the error text
1045 * might go stale if there are any other intervening rtas calls,
1046 * this routine must be called atomically with whatever produced
1047 * the error (i.e. with rtas_lock still held from the previous call).
1049 static char *__fetch_rtas_last_error(char *altbuf)
1051 const s32 token = rtas_function_token(RTAS_FN_RTAS_LAST_ERROR);
1052 struct rtas_args err_args, save_args;
1056 lockdep_assert_held(&rtas_lock);
1061 bufsz = rtas_get_error_log_max();
1063 err_args.token = cpu_to_be32(token);
1064 err_args.nargs = cpu_to_be32(2);
1065 err_args.nret = cpu_to_be32(1);
1066 err_args.args[0] = cpu_to_be32(__pa(rtas_err_buf));
1067 err_args.args[1] = cpu_to_be32(bufsz);
1068 err_args.args[2] = 0;
1070 save_args = rtas_args;
1071 rtas_args = err_args;
1073 do_enter_rtas(&rtas_args);
1075 err_args = rtas_args;
1076 rtas_args = save_args;
1078 /* Log the error in the unlikely case that there was one. */
1079 if (unlikely(err_args.args[2] == 0)) {
1084 if (slab_is_available())
1085 buf = kmalloc(RTAS_ERROR_LOG_MAX, GFP_ATOMIC);
1088 memmove(buf, rtas_err_buf, RTAS_ERROR_LOG_MAX);
1094 #define get_errorlog_buffer() kmalloc(RTAS_ERROR_LOG_MAX, GFP_KERNEL)
1096 #else /* CONFIG_RTAS_ERROR_LOGGING */
1097 #define __fetch_rtas_last_error(x) NULL
1098 #define get_errorlog_buffer() NULL
1099 static void __init init_error_log_max(void) {}
1104 va_rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret,
1109 args->token = cpu_to_be32(token);
1110 args->nargs = cpu_to_be32(nargs);
1111 args->nret = cpu_to_be32(nret);
1112 args->rets = &(args->args[nargs]);
1114 for (i = 0; i < nargs; ++i)
1115 args->args[i] = cpu_to_be32(va_arg(list, __u32));
1117 for (i = 0; i < nret; ++i)
1120 do_enter_rtas(args);
1124 * rtas_call_unlocked() - Invoke an RTAS firmware function without synchronization.
1125 * @args: RTAS parameter block to be used for the call, must obey RTAS addressing
1127 * @token: Identifies the function being invoked.
1128 * @nargs: Number of input parameters. Does not include token.
1129 * @nret: Number of output parameters, including the call status.
1130 * @....: List of @nargs input parameters.
1132 * Invokes the RTAS function indicated by @token, which the caller
1133 * should obtain via rtas_function_token().
1135 * This function is similar to rtas_call(), but must be used with a
1136 * limited set of RTAS calls specifically exempted from the general
1137 * requirement that only one RTAS call may be in progress at any
1138 * time. Examples include stop-self and ibm,nmi-interlock.
1140 void rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret, ...)
1144 va_start(list, nret);
1145 va_rtas_call_unlocked(args, token, nargs, nret, list);
1149 static bool token_is_restricted_errinjct(s32 token)
1151 return token == rtas_function_token(RTAS_FN_IBM_OPEN_ERRINJCT) ||
1152 token == rtas_function_token(RTAS_FN_IBM_ERRINJCT);
1156 * rtas_call() - Invoke an RTAS firmware function.
1157 * @token: Identifies the function being invoked.
1158 * @nargs: Number of input parameters. Does not include token.
1159 * @nret: Number of output parameters, including the call status.
1160 * @outputs: Array of @nret output words.
1161 * @....: List of @nargs input parameters.
1163 * Invokes the RTAS function indicated by @token, which the caller
1164 * should obtain via rtas_function_token().
1166 * The @nargs and @nret arguments must match the number of input and
1167 * output parameters specified for the RTAS function.
1169 * rtas_call() returns RTAS status codes, not conventional Linux errno
1170 * values. Callers must translate any failure to an appropriate errno
1171 * in syscall context. Most callers of RTAS functions that can return
1172 * -2 or 990x should use rtas_busy_delay() to correctly handle those
1173 * statuses before calling again.
1175 * The return value descriptions are adapted from 7.2.8 [RTAS] Return
1176 * Codes of the PAPR and CHRP specifications.
1178 * Context: Process context preferably, interrupt context if
1179 * necessary. Acquires an internal spinlock and may perform
1180 * GFP_ATOMIC slab allocation in error path. Unsafe for NMI
1183 * * 0 - RTAS function call succeeded.
1184 * * -1 - RTAS function encountered a hardware or
1185 * platform error, or the token is invalid,
1186 * or the function is restricted by kernel policy.
1187 * * -2 - Specs say "A necessary hardware device was busy,
1188 * and the requested function could not be
1189 * performed. The operation should be retried at
1190 * a later time." This is misleading, at least with
1191 * respect to current RTAS implementations. What it
1192 * usually means in practice is that the function
1193 * could not be completed while meeting RTAS's
1194 * deadline for returning control to the OS (250us
1195 * for PAPR/PowerVM, typically), but the call may be
1196 * immediately reattempted to resume work on it.
1197 * * -3 - Parameter error.
1198 * * -7 - Unexpected state change.
1199 * * 9000...9899 - Vendor-specific success codes.
1200 * * 9900...9905 - Advisory extended delay. Caller should try
1201 * again after ~10^x ms has elapsed, where x is
1202 * the last digit of the status [0-5]. Again going
1203 * beyond the PAPR text, 990x on PowerVM indicates
1204 * contention for RTAS-internal resources. Other
1205 * RTAS call sequences in progress should be
1206 * allowed to complete before reattempting the
1208 * * -9000 - Multi-level isolation error.
1209 * * -9999...-9004 - Vendor-specific error codes.
1210 * * Additional negative values - Function-specific error.
1211 * * Additional positive values - Function-specific success.
1213 int rtas_call(int token, int nargs, int nret, int *outputs, ...)
1215 struct pin_cookie cookie;
1218 unsigned long flags;
1219 struct rtas_args *args;
1220 char *buff_copy = NULL;
1223 if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
1226 if (token_is_restricted_errinjct(token)) {
1228 * It would be nicer to not discard the error value
1229 * from security_locked_down(), but callers expect an
1230 * RTAS status, not an errno.
1232 if (security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION))
1236 if ((mfmsr() & (MSR_IR|MSR_DR)) != (MSR_IR|MSR_DR)) {
1241 raw_spin_lock_irqsave(&rtas_lock, flags);
1242 cookie = lockdep_pin_lock(&rtas_lock);
1244 /* We use the global rtas args buffer */
1247 va_start(list, outputs);
1248 va_rtas_call_unlocked(args, token, nargs, nret, list);
1251 /* A -1 return code indicates that the last command couldn't
1252 be completed due to a hardware error. */
1253 if (be32_to_cpu(args->rets[0]) == -1)
1254 buff_copy = __fetch_rtas_last_error(NULL);
1256 if (nret > 1 && outputs != NULL)
1257 for (i = 0; i < nret-1; ++i)
1258 outputs[i] = be32_to_cpu(args->rets[i + 1]);
1259 ret = (nret > 0) ? be32_to_cpu(args->rets[0]) : 0;
1261 lockdep_unpin_lock(&rtas_lock, cookie);
1262 raw_spin_unlock_irqrestore(&rtas_lock, flags);
1265 log_error(buff_copy, ERR_TYPE_RTAS_LOG, 0);
1266 if (slab_is_available())
1271 EXPORT_SYMBOL_GPL(rtas_call);
1274 * rtas_busy_delay_time() - From an RTAS status value, calculate the
1275 * suggested delay time in milliseconds.
1277 * @status: a value returned from rtas_call() or similar APIs which return
1278 * the status of a RTAS function call.
1280 * Context: Any context.
1283 * * 100000 - If @status is 9905.
1284 * * 10000 - If @status is 9904.
1285 * * 1000 - If @status is 9903.
1286 * * 100 - If @status is 9902.
1287 * * 10 - If @status is 9901.
1288 * * 1 - If @status is either 9900 or -2. This is "wrong" for -2, but
1289 * some callers depend on this behavior, and the worst outcome
1290 * is that they will delay for longer than necessary.
1291 * * 0 - If @status is not a busy or extended delay value.
1293 unsigned int rtas_busy_delay_time(int status)
1296 unsigned int ms = 0;
1298 if (status == RTAS_BUSY) {
1300 } else if (status >= RTAS_EXTENDED_DELAY_MIN &&
1301 status <= RTAS_EXTENDED_DELAY_MAX) {
1302 order = status - RTAS_EXTENDED_DELAY_MIN;
1303 for (ms = 1; order > 0; order--)
1311 * Early boot fallback for rtas_busy_delay().
1313 static bool __init rtas_busy_delay_early(int status)
1315 static size_t successive_ext_delays __initdata;
1319 case RTAS_EXTENDED_DELAY_MIN...RTAS_EXTENDED_DELAY_MAX:
1321 * In the unlikely case that we receive an extended
1322 * delay status in early boot, the OS is probably not
1323 * the cause, and there's nothing we can do to clear
1324 * the condition. Best we can do is delay for a bit
1325 * and hope it's transient. Lie to the caller if it
1326 * seems like we're stuck in a retry loop.
1330 successive_ext_delays += 1;
1331 if (successive_ext_delays > 1000) {
1332 pr_err("too many extended delays, giving up\n");
1335 successive_ext_delays = 0;
1340 successive_ext_delays = 0;
1344 successive_ext_delays = 0;
1352 * rtas_busy_delay() - helper for RTAS busy and extended delay statuses
1354 * @status: a value returned from rtas_call() or similar APIs which return
1355 * the status of a RTAS function call.
1357 * Context: Process context. May sleep or schedule.
1360 * * true - @status is RTAS_BUSY or an extended delay hint. The
1361 * caller may assume that the CPU has been yielded if necessary,
1362 * and that an appropriate delay for @status has elapsed.
1363 * Generally the caller should reattempt the RTAS call which
1366 * * false - @status is not @RTAS_BUSY nor an extended delay hint. The
1367 * caller is responsible for handling @status.
1369 bool __ref rtas_busy_delay(int status)
1375 * Can't do timed sleeps before timekeeping is up.
1377 if (system_state < SYSTEM_SCHEDULING)
1378 return rtas_busy_delay_early(status);
1381 case RTAS_EXTENDED_DELAY_MIN...RTAS_EXTENDED_DELAY_MAX:
1383 ms = rtas_busy_delay_time(status);
1385 * The extended delay hint can be as high as 100 seconds.
1386 * Surely any function returning such a status is either
1387 * buggy or isn't going to be significantly slowed by us
1388 * polling at 1HZ. Clamp the sleep time to one second.
1390 ms = clamp(ms, 1U, 1000U);
1392 * The delay hint is an order-of-magnitude suggestion, not
1393 * a minimum. It is fine, possibly even advantageous, for
1394 * us to pause for less time than hinted. For small values,
1395 * use usleep_range() to ensure we don't sleep much longer
1396 * than actually needed.
1398 * See Documentation/timers/timers-howto.rst for
1399 * explanation of the threshold used here. In effect we use
1400 * usleep_range() for 9900 and 9901, msleep() for
1404 usleep_range(ms * 100, ms * 1000);
1411 * We should call again immediately if there's no other
1419 * Not a busy or extended delay status; the caller should
1420 * handle @status itself. Ensure we warn on misuses in
1421 * atomic context regardless.
1429 EXPORT_SYMBOL_GPL(rtas_busy_delay);
1431 int rtas_error_rc(int rtas_rc)
1436 case RTAS_HARDWARE_ERROR: /* Hardware Error */
1439 case RTAS_INVALID_PARAMETER: /* Bad indicator/domain/etc */
1442 case -9000: /* Isolation error */
1445 case -9001: /* Outstanding TCE/PTE */
1448 case -9002: /* No usable slot */
1452 pr_err("%s: unexpected error %d\n", __func__, rtas_rc);
1458 EXPORT_SYMBOL_GPL(rtas_error_rc);
1460 int rtas_get_power_level(int powerdomain, int *level)
1462 int token = rtas_function_token(RTAS_FN_GET_POWER_LEVEL);
1465 if (token == RTAS_UNKNOWN_SERVICE)
1468 while ((rc = rtas_call(token, 1, 2, level, powerdomain)) == RTAS_BUSY)
1472 return rtas_error_rc(rc);
1475 EXPORT_SYMBOL_GPL(rtas_get_power_level);
1477 int rtas_set_power_level(int powerdomain, int level, int *setlevel)
1479 int token = rtas_function_token(RTAS_FN_SET_POWER_LEVEL);
1482 if (token == RTAS_UNKNOWN_SERVICE)
1486 rc = rtas_call(token, 2, 2, setlevel, powerdomain, level);
1487 } while (rtas_busy_delay(rc));
1490 return rtas_error_rc(rc);
1493 EXPORT_SYMBOL_GPL(rtas_set_power_level);
1495 int rtas_get_sensor(int sensor, int index, int *state)
1497 int token = rtas_function_token(RTAS_FN_GET_SENSOR_STATE);
1500 if (token == RTAS_UNKNOWN_SERVICE)
1504 rc = rtas_call(token, 2, 2, state, sensor, index);
1505 } while (rtas_busy_delay(rc));
1508 return rtas_error_rc(rc);
1511 EXPORT_SYMBOL_GPL(rtas_get_sensor);
1513 int rtas_get_sensor_fast(int sensor, int index, int *state)
1515 int token = rtas_function_token(RTAS_FN_GET_SENSOR_STATE);
1518 if (token == RTAS_UNKNOWN_SERVICE)
1521 rc = rtas_call(token, 2, 2, state, sensor, index);
1522 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
1523 rc <= RTAS_EXTENDED_DELAY_MAX));
1526 return rtas_error_rc(rc);
1530 bool rtas_indicator_present(int token, int *maxindex)
1532 int proplen, count, i;
1533 const struct indicator_elem {
1538 indicators = of_get_property(rtas.dev, "rtas-indicators", &proplen);
1542 count = proplen / sizeof(struct indicator_elem);
1544 for (i = 0; i < count; i++) {
1545 if (__be32_to_cpu(indicators[i].token) != token)
1548 *maxindex = __be32_to_cpu(indicators[i].maxindex);
1555 int rtas_set_indicator(int indicator, int index, int new_value)
1557 int token = rtas_function_token(RTAS_FN_SET_INDICATOR);
1560 if (token == RTAS_UNKNOWN_SERVICE)
1564 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
1565 } while (rtas_busy_delay(rc));
1568 return rtas_error_rc(rc);
1571 EXPORT_SYMBOL_GPL(rtas_set_indicator);
1574 * Ignoring RTAS extended delay
1576 int rtas_set_indicator_fast(int indicator, int index, int new_value)
1578 int token = rtas_function_token(RTAS_FN_SET_INDICATOR);
1581 if (token == RTAS_UNKNOWN_SERVICE)
1584 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
1586 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
1587 rc <= RTAS_EXTENDED_DELAY_MAX));
1590 return rtas_error_rc(rc);
1596 * rtas_ibm_suspend_me() - Call ibm,suspend-me to suspend the LPAR.
1598 * @fw_status: RTAS call status will be placed here if not NULL.
1600 * rtas_ibm_suspend_me() should be called only on a CPU which has
1601 * received H_CONTINUE from the H_JOIN hcall. All other active CPUs
1602 * should be waiting to return from H_JOIN.
1604 * rtas_ibm_suspend_me() may suspend execution of the OS
1605 * indefinitely. Callers should take appropriate measures upon return, such as
1606 * resetting watchdog facilities.
1608 * Callers may choose to retry this call if @fw_status is
1609 * %RTAS_THREADS_ACTIVE.
1612 * 0 - The partition has resumed from suspend, possibly after
1613 * migration to a different host.
1614 * -ECANCELED - The operation was aborted.
1615 * -EAGAIN - There were other CPUs not in H_JOIN at the time of the call.
1616 * -EBUSY - Some other condition prevented the suspend from succeeding.
1617 * -EIO - Hardware/platform error.
1619 int rtas_ibm_suspend_me(int *fw_status)
1621 int token = rtas_function_token(RTAS_FN_IBM_SUSPEND_ME);
1625 fwrc = rtas_call(token, 0, 1, NULL);
1631 case RTAS_SUSPEND_ABORTED:
1634 case RTAS_THREADS_ACTIVE:
1637 case RTAS_NOT_SUSPENDABLE:
1638 case RTAS_OUTSTANDING_COPROC:
1653 void __noreturn rtas_restart(char *cmd)
1655 if (rtas_flash_term_hook)
1656 rtas_flash_term_hook(SYS_RESTART);
1657 pr_emerg("system-reboot returned %d\n",
1658 rtas_call(rtas_function_token(RTAS_FN_SYSTEM_REBOOT), 0, 1, NULL));
1662 void rtas_power_off(void)
1664 if (rtas_flash_term_hook)
1665 rtas_flash_term_hook(SYS_POWER_OFF);
1666 /* allow power on only with power button press */
1667 pr_emerg("power-off returned %d\n",
1668 rtas_call(rtas_function_token(RTAS_FN_POWER_OFF), 2, 1, NULL, -1, -1));
1672 void __noreturn rtas_halt(void)
1674 if (rtas_flash_term_hook)
1675 rtas_flash_term_hook(SYS_HALT);
1676 /* allow power on only with power button press */
1677 pr_emerg("power-off returned %d\n",
1678 rtas_call(rtas_function_token(RTAS_FN_POWER_OFF), 2, 1, NULL, -1, -1));
1682 /* Must be in the RMO region, so we place it here */
1683 static char rtas_os_term_buf[2048];
1684 static bool ibm_extended_os_term;
1686 void rtas_os_term(char *str)
1688 s32 token = rtas_function_token(RTAS_FN_IBM_OS_TERM);
1689 static struct rtas_args args;
1693 * Firmware with the ibm,extended-os-term property is guaranteed
1694 * to always return from an ibm,os-term call. Earlier versions without
1695 * this property may terminate the partition which we want to avoid
1696 * since it interferes with panic_timeout.
1699 if (token == RTAS_UNKNOWN_SERVICE || !ibm_extended_os_term)
1702 snprintf(rtas_os_term_buf, 2048, "OS panic: %s", str);
1705 * Keep calling as long as RTAS returns a "try again" status,
1706 * but don't use rtas_busy_delay(), which potentially
1710 rtas_call_unlocked(&args, token, 1, 1, NULL, __pa(rtas_os_term_buf));
1711 status = be32_to_cpu(args.rets[0]);
1712 } while (rtas_busy_delay_time(status));
1715 pr_emerg("ibm,os-term call failed %d\n", status);
1719 * rtas_activate_firmware() - Activate a new version of firmware.
1721 * Context: This function may sleep.
1723 * Activate a new version of partition firmware. The OS must call this
1724 * after resuming from a partition hibernation or migration in order
1725 * to maintain the ability to perform live firmware updates. It's not
1726 * catastrophic for this method to be absent or to fail; just log the
1727 * condition in that case.
1729 void rtas_activate_firmware(void)
1731 int token = rtas_function_token(RTAS_FN_IBM_ACTIVATE_FIRMWARE);
1734 if (token == RTAS_UNKNOWN_SERVICE) {
1735 pr_notice("ibm,activate-firmware method unavailable\n");
1739 mutex_lock(&rtas_ibm_activate_firmware_lock);
1742 fwrc = rtas_call(token, 0, 1, NULL);
1743 } while (rtas_busy_delay(fwrc));
1745 mutex_unlock(&rtas_ibm_activate_firmware_lock);
1748 pr_err("ibm,activate-firmware failed (%i)\n", fwrc);
1752 * get_pseries_errorlog() - Find a specific pseries error log in an RTAS
1753 * extended event log.
1754 * @log: RTAS error/event log
1755 * @section_id: two character section identifier
1757 * Return: A pointer to the specified errorlog or NULL if not found.
1759 noinstr struct pseries_errorlog *get_pseries_errorlog(struct rtas_error_log *log,
1760 uint16_t section_id)
1762 struct rtas_ext_event_log_v6 *ext_log =
1763 (struct rtas_ext_event_log_v6 *)log->buffer;
1764 struct pseries_errorlog *sect;
1765 unsigned char *p, *log_end;
1766 uint32_t ext_log_length = rtas_error_extended_log_length(log);
1767 uint8_t log_format = rtas_ext_event_log_format(ext_log);
1768 uint32_t company_id = rtas_ext_event_company_id(ext_log);
1770 /* Check that we understand the format */
1771 if (ext_log_length < sizeof(struct rtas_ext_event_log_v6) ||
1772 log_format != RTAS_V6EXT_LOG_FORMAT_EVENT_LOG ||
1773 company_id != RTAS_V6EXT_COMPANY_ID_IBM)
1776 log_end = log->buffer + ext_log_length;
1777 p = ext_log->vendor_log;
1779 while (p < log_end) {
1780 sect = (struct pseries_errorlog *)p;
1781 if (pseries_errorlog_id(sect) == section_id)
1783 p += pseries_errorlog_length(sect);
1790 * The sys_rtas syscall, as originally designed, allows root to pass
1791 * arbitrary physical addresses to RTAS calls. A number of RTAS calls
1792 * can be abused to write to arbitrary memory and do other things that
1793 * are potentially harmful to system integrity, and thus should only
1794 * be used inside the kernel and not exposed to userspace.
1796 * All known legitimate users of the sys_rtas syscall will only ever
1797 * pass addresses that fall within the RMO buffer, and use a known
1798 * subset of RTAS calls.
1800 * Accordingly, we filter RTAS requests to check that the call is
1801 * permitted, and that provided pointers fall within the RMO buffer.
1802 * If a function is allowed to be invoked via the syscall, then its
1803 * entry in the rtas_functions table points to a rtas_filter that
1804 * describes its constraints, with the indexes of the parameters which
1805 * are expected to contain addresses and sizes of buffers allocated
1806 * inside the RMO buffer.
1809 static bool in_rmo_buf(u32 base, u32 end)
1811 return base >= rtas_rmo_buf &&
1812 base < (rtas_rmo_buf + RTAS_USER_REGION_SIZE) &&
1814 end >= rtas_rmo_buf &&
1815 end < (rtas_rmo_buf + RTAS_USER_REGION_SIZE);
1818 static bool block_rtas_call(const struct rtas_function *func, int nargs,
1819 struct rtas_args *args)
1821 const struct rtas_filter *f;
1822 const bool is_platform_dump =
1823 func == &rtas_function_table[RTAS_FNIDX__IBM_PLATFORM_DUMP];
1824 const bool is_config_conn =
1825 func == &rtas_function_table[RTAS_FNIDX__IBM_CONFIGURE_CONNECTOR];
1826 u32 base, size, end;
1829 * Only functions with filters attached are allowed.
1835 * And some functions aren't allowed on LE.
1837 if (IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) && func->banned_for_syscall_on_le)
1840 if (f->buf_idx1 != -1) {
1841 base = be32_to_cpu(args->args[f->buf_idx1]);
1842 if (f->size_idx1 != -1)
1843 size = be32_to_cpu(args->args[f->size_idx1]);
1844 else if (f->fixed_size)
1845 size = f->fixed_size;
1849 end = base + size - 1;
1852 * Special case for ibm,platform-dump - NULL buffer
1853 * address is used to indicate end of dump processing
1855 if (is_platform_dump && base == 0)
1858 if (!in_rmo_buf(base, end))
1862 if (f->buf_idx2 != -1) {
1863 base = be32_to_cpu(args->args[f->buf_idx2]);
1864 if (f->size_idx2 != -1)
1865 size = be32_to_cpu(args->args[f->size_idx2]);
1866 else if (f->fixed_size)
1867 size = f->fixed_size;
1870 end = base + size - 1;
1873 * Special case for ibm,configure-connector where the
1876 if (is_config_conn && base == 0)
1879 if (!in_rmo_buf(base, end))
1885 pr_err_ratelimited("sys_rtas: RTAS call blocked - exploit attempt?\n");
1886 pr_err_ratelimited("sys_rtas: %s nargs=%d (called by %s)\n",
1887 func->name, nargs, current->comm);
1891 /* We assume to be passed big endian arguments */
1892 SYSCALL_DEFINE1(rtas, struct rtas_args __user *, uargs)
1894 const struct rtas_function *func;
1895 struct pin_cookie cookie;
1896 struct rtas_args args;
1897 unsigned long flags;
1898 char *buff_copy, *errbuf = NULL;
1899 int nargs, nret, token;
1901 if (!capable(CAP_SYS_ADMIN))
1907 if (copy_from_user(&args, uargs, 3 * sizeof(u32)) != 0)
1910 nargs = be32_to_cpu(args.nargs);
1911 nret = be32_to_cpu(args.nret);
1912 token = be32_to_cpu(args.token);
1914 if (nargs >= ARRAY_SIZE(args.args)
1915 || nret > ARRAY_SIZE(args.args)
1916 || nargs + nret > ARRAY_SIZE(args.args))
1920 if (copy_from_user(args.args, uargs->args,
1921 nargs * sizeof(rtas_arg_t)) != 0)
1925 * If this token doesn't correspond to a function the kernel
1926 * understands, you're not allowed to call it.
1928 func = rtas_token_to_function_untrusted(token);
1932 args.rets = &args.args[nargs];
1933 memset(args.rets, 0, nret * sizeof(rtas_arg_t));
1935 if (block_rtas_call(func, nargs, &args))
1938 if (token_is_restricted_errinjct(token)) {
1941 err = security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION);
1946 /* Need to handle ibm,suspend_me call specially */
1947 if (token == rtas_function_token(RTAS_FN_IBM_SUSPEND_ME)) {
1950 * rtas_ibm_suspend_me assumes the streamid handle is in cpu
1951 * endian, or at least the hcall within it requires it.
1954 u64 handle = ((u64)be32_to_cpu(args.args[0]) << 32)
1955 | be32_to_cpu(args.args[1]);
1956 rc = rtas_syscall_dispatch_ibm_suspend_me(handle);
1958 args.rets[0] = cpu_to_be32(RTAS_NOT_SUSPENDABLE);
1959 else if (rc == -EIO)
1960 args.rets[0] = cpu_to_be32(-1);
1966 buff_copy = get_errorlog_buffer();
1969 * If this function has a mutex assigned to it, we must
1970 * acquire it to avoid interleaving with any kernel-based uses
1971 * of the same function. Kernel-based sequences acquire the
1972 * appropriate mutex explicitly.
1975 mutex_lock(func->lock);
1977 raw_spin_lock_irqsave(&rtas_lock, flags);
1978 cookie = lockdep_pin_lock(&rtas_lock);
1981 do_enter_rtas(&rtas_args);
1984 /* A -1 return code indicates that the last command couldn't
1985 be completed due to a hardware error. */
1986 if (be32_to_cpu(args.rets[0]) == -1)
1987 errbuf = __fetch_rtas_last_error(buff_copy);
1989 lockdep_unpin_lock(&rtas_lock, cookie);
1990 raw_spin_unlock_irqrestore(&rtas_lock, flags);
1993 mutex_unlock(func->lock);
1997 log_error(errbuf, ERR_TYPE_RTAS_LOG, 0);
2002 /* Copy out args. */
2003 if (copy_to_user(uargs->args + nargs,
2005 nret * sizeof(rtas_arg_t)) != 0)
2011 static void __init rtas_function_table_init(void)
2013 struct property *prop;
2015 for (size_t i = 0; i < ARRAY_SIZE(rtas_function_table); ++i) {
2016 struct rtas_function *curr = &rtas_function_table[i];
2017 struct rtas_function *prior;
2020 curr->token = RTAS_UNKNOWN_SERVICE;
2025 * Ensure table is sorted correctly for binary search
2026 * on function names.
2028 prior = &rtas_function_table[i - 1];
2030 cmp = strcmp(prior->name, curr->name);
2035 pr_err("'%s' has duplicate function table entries\n",
2038 pr_err("function table unsorted: '%s' wrongly precedes '%s'\n",
2039 prior->name, curr->name);
2043 for_each_property_of_node(rtas.dev, prop) {
2044 struct rtas_function *func;
2046 if (prop->length != sizeof(u32))
2049 func = __rtas_name_to_function(prop->name);
2053 func->token = be32_to_cpup((__be32 *)prop->value);
2055 pr_debug("function %s has token %u\n", func->name, func->token);
2060 * Call early during boot, before mem init, to retrieve the RTAS
2061 * information from the device-tree and allocate the RMO buffer for userland
2064 void __init rtas_initialize(void)
2066 unsigned long rtas_region = RTAS_INSTANTIATE_MAX;
2067 u32 base, size, entry;
2068 int no_base, no_size, no_entry;
2070 /* Get RTAS dev node and fill up our "rtas" structure with infos
2073 rtas.dev = of_find_node_by_name(NULL, "rtas");
2077 no_base = of_property_read_u32(rtas.dev, "linux,rtas-base", &base);
2078 no_size = of_property_read_u32(rtas.dev, "rtas-size", &size);
2079 if (no_base || no_size) {
2080 of_node_put(rtas.dev);
2087 no_entry = of_property_read_u32(rtas.dev, "linux,rtas-entry", &entry);
2088 rtas.entry = no_entry ? rtas.base : entry;
2090 init_error_log_max();
2092 /* Must be called before any function token lookups */
2093 rtas_function_table_init();
2096 * Discover this now to avoid a device tree lookup in the
2099 ibm_extended_os_term = of_property_read_bool(rtas.dev, "ibm,extended-os-term");
2101 /* If RTAS was found, allocate the RMO buffer for it and look for
2102 * the stop-self token if any
2105 if (firmware_has_feature(FW_FEATURE_LPAR))
2106 rtas_region = min(ppc64_rma_size, RTAS_INSTANTIATE_MAX);
2108 rtas_rmo_buf = memblock_phys_alloc_range(RTAS_USER_REGION_SIZE, PAGE_SIZE,
2111 panic("ERROR: RTAS: Failed to allocate %lx bytes below %pa\n",
2112 PAGE_SIZE, &rtas_region);
2114 rtas_work_area_reserve_arena(rtas_region);
2117 int __init early_init_dt_scan_rtas(unsigned long node,
2118 const char *uname, int depth, void *data)
2120 const u32 *basep, *entryp, *sizep;
2122 if (depth != 1 || strcmp(uname, "rtas") != 0)
2125 basep = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
2126 entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
2127 sizep = of_get_flat_dt_prop(node, "rtas-size", NULL);
2130 /* need this feature to decide the crashkernel offset */
2131 if (of_get_flat_dt_prop(node, "ibm,hypertas-functions", NULL))
2132 powerpc_firmware_features |= FW_FEATURE_LPAR;
2135 if (basep && entryp && sizep) {
2137 rtas.entry = *entryp;
2141 #ifdef CONFIG_UDBG_RTAS_CONSOLE
2142 basep = of_get_flat_dt_prop(node, "put-term-char", NULL);
2144 rtas_putchar_token = *basep;
2146 basep = of_get_flat_dt_prop(node, "get-term-char", NULL);
2148 rtas_getchar_token = *basep;
2150 if (rtas_putchar_token != RTAS_UNKNOWN_SERVICE &&
2151 rtas_getchar_token != RTAS_UNKNOWN_SERVICE)
2152 udbg_init_rtas_console();
2160 static DEFINE_RAW_SPINLOCK(timebase_lock);
2161 static u64 timebase = 0;
2163 void rtas_give_timebase(void)
2165 unsigned long flags;
2167 raw_spin_lock_irqsave(&timebase_lock, flags);
2169 rtas_call(rtas_function_token(RTAS_FN_FREEZE_TIME_BASE), 0, 1, NULL);
2170 timebase = get_tb();
2171 raw_spin_unlock(&timebase_lock);
2175 rtas_call(rtas_function_token(RTAS_FN_THAW_TIME_BASE), 0, 1, NULL);
2176 local_irq_restore(flags);
2179 void rtas_take_timebase(void)
2183 raw_spin_lock(&timebase_lock);
2184 set_tb(timebase >> 32, timebase & 0xffffffff);
2186 raw_spin_unlock(&timebase_lock);